Uteke CLI
Local memory for agents in a single Rust binary: stores, searches by meaning and keyword, works with no network and no keys
- 278 ★ on the repository
- Last commit
- Apache-2.0
Extended access
Why this level
- Keeps everything the agent remembered on disk, and notes may include secrets
- Server mode listens on port 8767, and launching it on 0.0.0.0 opens it to the network
- The first run downloads a model from an external address and runs it locally
Install
- Works in
Manual install
curl -sSL codecora.dev/uteke/install | shWorks on macOS, Linux and Windows. Alternatives: `brew install codecoradev/tap/uteke`, `cargo install uteke-cli`, the Docker image `ghcr.io/codecoradev/uteke:latest` and GitHub releases.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Keeps everything the agent remembered on disk, and notes may include secrets
- Server mode listens on port 8767, and launching it on 0.0.0.0 opens it to the network
- The first run downloads a model from an external address and runs it locally
This is third-party code. Review the repository files before installing.
What it does
Uteke is a Rust command-line program that keeps an AI agent's memory on your computer. `uteke remember` stores a note and `uteke recall` finds it by meaning and by exact words. Search merges an HNSW vector index with SQLite FTS5 full-text search, and a local ONNX model computes the vectors, so no keys or network are needed after the first model download. Claude Code and Cursor get an MCP server, `uteke-mcp`. Rooms are shared by several agents: notes are grouped by topic, each has an author, and `uteke recall --room` searches the shared history. A note can be searched as it was on a date (`--at`), linked to others, merged with duplicates and given importance and a decay period. Server mode keeps the model in memory so calls skip the startup cost; a Docker image and read-only tokens exist. Retrieval makes no model calls, and the author published the LongMemEval-S results and scoring scripts in the repository.
Who it is for. For developers and teams whose agent forgets agreements between sessions and who need memory on their own machine with no cloud.
Good fit when
- An agent forgets decisions between sessions and memory must stay on your machine
- Several agents work on one project and need a room with note authorship
- You need to see what an agent knew on a given date and return to earlier notes
- A folder of notes should be loaded as a knowledge base with a batch import
- Data cannot go to the cloud: healthcare, finance, legal
Not a fit when
- There is no room for the model: the first embedding download is about 200 MB
- You need a ready cloud memory with browser sharing: Uteke runs locally or on your own server
- The agent needs only a short project note: a plain rules file is enough
- Background servers are forbidden on the machine: server mode runs as a persistent process
Example request
Remember in Uteke that release 2.1 goes to staging at 3pm, and tomorrow find it with the question "when do we deploy"Limitations
The first run downloads an embedding model of about 200 MB, after which it works offline. Data lives in `~/.codecora/uteke/`. The author measured LongMemEval-S results themselves and published scripts to repeat it; a rerun costs roughly 10 dollars. The strict recall_all@5 is 88.0%: ordering suffers when a question needs several sessions at once. Homebrew install is available on Apple Silicon macOS and Linux; other systems use `cargo`, Docker or prebuilt binaries.
How to disable. Remove the `uteke` entry from `.mcp.json`, stop the server if it runs and delete the binary. To wipe memory completely, delete the `~/.codecora/uteke/` folder.
FAQ
How do I connect Uteke to Claude Code?
Add `{ "mcpServers": { "uteke": { "command": "uteke-mcp" } } }` to `.mcp.json`, or run `uteke onboard`: the wizard detects the agent and configures everything.
Does Uteke need API keys?
No. A local ONNX model of 768 dimensions computes the vectors and search runs in SQLite. No keys are needed and there is no telemetry or cloud.
What are rooms in Uteke?
A room groups notes by context: meetings, projects, clients. Several agents read and write to it, and each note carries its author. `uteke room create "engineering"` creates one.
How do I find a note as of a date?
Add the `--at` flag: `uteke recall "deploy" --at 2025-01-15` returns what was stored on that date.
How can I check the claimed search accuracy?
The whole harness is in the repository: public data, saved outputs for three releases and scoring scripts in `benchmarks/longmemeval/`. The guide is REPRODUCING.md. For your own volumes use `uteke bench`.
Discussion
No questions or reviews yet.
Related
CLI toolsA long-term memory MCP server for agents: one Rust binary that runs offline and never calls a language model
Long-term memory for agents: it distills knowledge from session history and brings it back in new tasks, one store shared across Claude Code, Codex, Cursor and
Shared local memory for AI agents: Claude Code, Codex, Cursor and others find the agreements of past sessions in it and record new ones
Local persistent memory for coding agents: hooks record every session, and Claude Code, Codex or Cursor recall past decisions through 54 MCP tools
Self-hosted long-term memory for agents: semantic search, a knowledge graph and autonomous consolidation over MCP and a REST API
A long-term memory MCP server for AI agents: facts fade along a forgetting curve and related records are compressed into a summary
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Uteke (this entry) | no | 278 | ||
| agentmemory | yes | not required | 29,985 | local memory with hooks that record every session |
| memsearch | no | 2,805 | shared semantic memory with conversations stored as markdown |
README in short
The README starts with installation and a two-command example, then shows LongMemEval-S results with tables by release and question category. It continues with a comparison to unnamed alternatives, the list of search, room, link and import features, a hybrid search diagram and Docker deployment modes.